Upload
others
View
2
Download
0
Embed Size (px)
Citation preview
transactions of theamerican mathematical societyVolume 330, Number 1, March 1992
A CHARACTERIZATION OF COCOMPACT HYPERBOLIC ANDFINITE-VOLUME HYPERBOLIC GROUPS IN DIMENSION THREE
J. W. CANNON AND DARYL COOPER
Abstract. We show that a cocompact hyperbolic group in dimension 3 is char-
acterized by certain properties of its word metric which depend only on the
group structure and not on any action on hyperbolic space. We prove a similar
theorem for finite-volume hyperbolic groups in dimension 3.
1. Introduction
A group G is called cocompact hyperbolic of dimension n if and only if itcan act isometrically, properly discontinuously, and cocompactly on hyperbolic
space H" . Proper discontinuity implies that the action has at most finite kernel.
Our main result states that a group G is cocompact hyperbolic of dimension
3 if and only if it has a finite generating set C and associated Cayley graph
F = r(Cr, C) such that the word metric on T is comparable in the large (quasi-
isometric) with the standard hyperbolic metric on H3. We have a similar result
for finite volume hyperbolic groups in dimension 3. We believe that our results
generalize to all dimensions, at least in the cocompact case. Our result for
groups of finite volume employs a different metric space. Instead of the Cayley
graph with word metric, we use an augmented Cayley graph associated with a
group G and a finite family of Euclidean subgroups of G. We prove that the
group acts isometrically and properly discontinuously on H3 with orbit space
of finite volume if and only if the augmented Cayley graph is quasi-isometric
with H3. Precise definitions of all of these terms will appear in §2. The proofof the cocompact case appears in §3; the proof of the finite volume case appears
in §4.Our results are similar to a result of Gromov [Gl] but without the added
assumption that the group be the fundamental group of a convex path space.
There are other results on the classification of groups up to quasi-isometry in[G2]. In particular a group is quasi-isometric to E" if and only if there is asubgroup Z" of finite index. It would be interesting to know which groups are
quasi-isometric to the 3-dimensional geometries H2 x E1, Solv, and PSL(2, R).
Gromov's work on groups with polynomial growth [G3, BK] implies that a group
is quasi-isometric to Nil if and only if it has a subgroup of finite index which
is the fundamental group of a compact Nil 3-orbifold. Notice that all compact
Received by the editors November 15, 1988 and, in revised form, January 22, 1990.
1980 Mathematics Subject Classification (1985 Revision). Primary 57M15, 30C60, 30F40.The authors were supported in part by grants from the National Science Foundation.
© 1992 American Mathematical Society0002-9947/92 $1.00+ $.25 per page
419
License or copyright restrictions may apply to redistribution; see http://www.ams.org/journal-terms-of-use
420 J. W. CANNON AND DARYL COOPER
spaces are quasi-isometric, so that all finite groups are quasi-isometric to S3.Further by the above result of Gromov the groups which are quasi-isometricto S2 x E1 are precisely those groups containing an infinite cyclic subgroup of
finite index.In [C] Cannon showed how to use the geometry of H" to give solutions
to the word problem and conjugacy problem for cocompact hyperbolic groups.
The results were exactly analogous to the original solutions by Dehn for sur-
face groups. Cannon proposes to show in another paper how to use that same
hyperbolic geometry, in conjunction with the results of this paper on groups of
finite volume, to establish analogous decision algorithms for hyperbolic groups
of finite volume.
2. Definitions and preliminary results
Here are the precise definitions. Each metric is a global topological metric
rather than an infinitesimal metric. All metrics are denoted by the symbol
d, perhaps with a subscript to indicate the space. If A is a set and e > 0,then N(A, e) denotes the e-neighborhood of A, that is the set of points at
distance < e from A . A group action of G on a metric space M with metric
d\t is a homomorphism from G into the homeomorphism group Homeo(M)
of M. The kernel of this homomorphism is called the kernel of the action.
The action is denoted in symbols by 4>: G x M -» M. The image 4>(g, x)
is denoted simply by g(x) or gx. A homeomorphism is an isometry if it
preserves the distance between points. The action is isometric if the image of
G under this homomorphism lies in the subgroup of isometries of the metricspace (M, du) • The action GxM -> M is said to be properly discontinuous if,
for each compact subset K of M, the set {g e G\K n g(K) ± 0} is finite. Fora properly discontinuous action, it is clear that the kernel of the homomorphismG —> VLomeo(M) is finite. The set of orbits M/G can be given the identification
topology. The resulting space is called the orbit space of the action. The action
is called cocompact if the orbit space is compact. In the case where M is
hyperbolic space H" and the action is properly discontinuous, then the orbit
space W/G inherits a natural volume element from H" . If the total volume
of H"/G is finite, then the action is said to be of finite volume. A group is
called finite-volume hyperbolic if it admits an action on H" that is isometric,
properly discontinuous, and of finite volume.
The natural setting for our results is that of geometric group theory which
studies a group by means of the geometries on which it acts isometrically. A
geometry is a metric space M with path metric d\¡ such that each closed metric
ball is compact. A group can act isometrically, properly discontinuously, and
cocompactly on some geometry if and only if it is finitely generated. (See [C2].)
If a group G is finitely generated, the Cayley group graph with respect to
any finite generating set is a geometry on which the group can act geometrically.Let G denote a group with generating set C = C~x. The Cayley group graph
T = T(G, C) of G with respect to C is a directed graph with vertex set G and
with directed edges of the form e = (v, c, v • c), where v e G and v - c e G
are vertices and c e C is a generator. The directed edges e = (v, c, v ■ c)
and e~x = (v • c, c~x, v) represent the same undirected edge of T and are
considered inverses of each other. The group graph just defined has a natural
License or copyright restrictions may apply to redistribution; see http://www.ams.org/journal-terms-of-use
HYPERBOLIC GROUPS 421
path metric called the word metric which assigns to each undirected edge the
path length of 1. The distance between the identity vertex and the vertex g of
r is then the minimal number of edges needed to join the two vertices. Thisdistance is, of course, the minimal word length of a word in the generators C
representing the element g. The closed metric balls are compact so that Y is
a geometry precisely in the case where the generating set is finite. The group G
acts isometrically on the group graph T = T(G, C) by left multiplication. Leftmultiplication by g e G takes the vertex v e G to the vertex g • v and the
edge e = (v , c, v • c) to the edge (g • v, c, g • v • c). If the generating set C isfinite, then this isometric action is also properly discontinuous and cocompact.
We call an action geometric if the action is an isometric, properly discon-
tinuous, and cocompact action on a geometry. A finitely generated group can,
of course, act geometrically on many different geometries. However, any two
geometries on which a finitely generated group acts geometrically are equivalentunder an equivalence relation which we shall call quasi-isometry. This relation-ship is also known as pseudo-isometry. We now explain this relationship. Let
(M, diu) and (N, du) be geometries. A relation (= multi-valued function)
R: M —► N is said to be K-quasi-Lipschitz, K>\,if
(i) R is everywhere defined (i.e., R(x) ^ 0 for each x G M) ; and
(ii) for each AcM, diam^(^) < K(di&mA + 1).
That is, the images of small sets must have uniformly bounded diameters, while
the images of large sets must have diameters satisfying a Lipschitz condition.
"Quasi" (= to some extent), as we use it here, is simply a substitute for "in the
large." Relations R: M -+ N and S: N —> M are K-quasi-inverses if each is
iT-quasi-Lipschitz, d/u(S°R, id^) < K, and dx(RoS, idyy) < K. A ÄT-quasi-Lipschitz relation R: M -» N is a K-quasi-isometry if it has a íT-quasi-inverse
S: N —> M. Given any two quasi-isometries, we may assume, if convenient,
that they each are assigned the same parameter K ; indeed, it suffices to use
the larger of given parameters for each. It is an easy exercise to prove that the
composition of two iT-quasi-isometries is a (K2 + Ä^-quasi-isometry.
We recall the following easy and well-known theorem.
Equivalence Theorem. If a group acts geometrically and cocompactly on two
geometries (M, d\f) and (N,d^), then (M, d¡u) and (N, d^) are quasi-isometric.
The proof is standard. The reader may construct a proof or consult [C2].
As a start, here are adequate definitions for R and S. Fix e > 0. Let Uo
and Vq be bounded, connected open sets in M and N, respectively, such that
the G-translates of Uo and Vo cover M and N, respectively. Let U and V
denote the e-neighborhoods of Uo and V0, respectively. Define R: M -> N
and S: N ^ M by
R(x) = {yeN\3geG,xegU,yegV}, and
S(y) = {xe M\3g eG,xegU,yegV}.
A notion which plays an essential role in our proofs is that of quasi-geodesic.
There are many equivalent formulations. The version that seems easiest to usein our case is the following. Let / denote an interval, finite or infinite, in the
space Z of integers with metric induced from the standard metric on the real
License or copyright restrictions may apply to redistribution; see http://www.ams.org/journal-terms-of-use
422 J. W. CANNON AND DARYL COOPER
numbers. Let M denote a metric space with metric d . A function /: / —> M
is said to be a K-quasi-geodesic if for each Ac I and B c f(I)
diam/(^)<i«:(diam^ + l) and diam/"1^) < Ä"(diamß + 1).
That is, / is a ÄNquasi-isometry onto its image, with / chosen to be a function
for convenience.
We obtain quasi-geodesics by means of two standard constructions. The
first starts with a minimizing geodesic g: [a, b] —*• M from a real interval
[a, b], finite or infinite, into a space M with path metric dut ■ One defines
/ = Zn[a,b] and / = g\I. Then / is a 1-quasi-geodesic. The secondconstruction starts with a K-quasi-geodesic /: / —> M and a A^-quasi-isometry
R: M —> N with AT-quasi-inverse S : N —> M. Let /':/—» TV be any functioncontained in the composite relation iîo/. An easy argument shows that /' is
a AT'-quasi-geodesic, with K' = 3K2 + K. Here is a key lemma.
Lemma. Let T denote a locally finite graph and (¡>: Y —► H3 a K-quasi-isometry.
Then there is a constant W(K) > 0, depending only on K, having the following
property. Let x, y e H3. Let x', y' e Y be points such that dn(x, <f>(x')) < K
and dfí(y, cj)(y')) < K. Let [x, y] denote the hyperbolic geodesic segment join-
ing x and y. Let a(x', y') denote a graph geodesic segment in Y connecting
x' and y'. Then (j)(a(x', y')) lies in the W(K) neighborhood of [x, y] in H3.
Proof. Let Xo,..., xn, Xo = x', x„ = y', be points of a(x', y') appear-
ing in the order named on a(x', y') with dr(x¡-X, x.) = 1 for i < n and
dr(xn-X, x„) < 1. Let /: {x0, ... , x„} -> H3 be any function contained
in (¡)\{xo, ... , xn} . As noted above, <?!>|{xo, ... , x„_i} is a (3K2 + AT)-quasi-
geodesic in H3. By [C, Theorem 2] there is a W0 depending only on 3K2 +
K such that /({xo, ... , x„__}) lies in the W0 neighborhood of the hyper-
bolic geodesic segment [/(xo), /(x„_i)] joining /(xo) and /(x„_i). But[f(xo) > f(xn-i)] lies in the 3K neighborhood of [x, y]. Hence we may takeW(K) = W0 + 3K.
Corollary. Let T denote a locally finite graph and </> : T —» H3 a K-quasi-
isometry. If a is a geodesic line in H3, then there is a geodesic line A in
T such that (¡)(A) lies within W(K) of a and a and 4>(A) have the same
endpoints at infinity in S___..
Proof. Let Ai, A2, ... denote geodesic segments in Y such that (j)(Ax ), (p(A2),
... approximate larger and larger geodesic segments in a as in the lemma.
Since each (j>(A¡) lies in the W(K) neighborhood of a and since only finitely
many segments of Y have images in the W(K) neighborhood of any given com-
pact segment of a, it follows that we may assume that the sequence A\, A2, ...
converges to a geodesic line A in Y which satisfies the conclusion of the corol-
lary.
3. Cocompact groups
With the Equivalence Theorem of §2 in mind, we might well discuss the main
theorem once more. A cocompact hyperbolic group acts geometrically on both
its Cayley graph and on hyperbolic space. Hence by the Equivalence Theorem
a necessary condition that a group be cocompact hyperbolic is that its Cayley
License or copyright restrictions may apply to redistribution; see http://www.ams.org/journal-terms-of-use
HYPERBOLIC GROUPS 423
graph be quasi-isometric with hyperbolic space. Our main theorem shows that
this necessary condition is also sufficient.
Theorem. Let G denote a group with finite generating set C = C~x. Let Y =
Y(G, C) denote the associated Cayley graph. Then G is cocompact hyperbolic
in dimension 3 if and only if Y is quasi-isometric with hyperbolic space H3.
Corollary. A group G is the fundamental group of some closed hyperbolic 3-
manifold if and only if G is finitely generated, torsion free, and the associated
Cayley graph is quasi-isometric with H3.
Proof. If G is cocompact hyperbolic in dimension 3, then, as noted in §1, the
group G is finitely generated and the associated Cayley graph is quasi-isometric
with hyperbolic space H3.
Suppose conversely that G has a finite generating set C = C~l with associ-
ated Cayley graph Y = Y(G, C) and that Y is quasi-isometric with H3. We
need to show that G is cocompact hyperbolic in dimension 3.
Here is an easy outline of the proof. Let 0 : Y -> H3 and y/ : H3 —> Y denote
ÄT-quasi-inverses. The formula gx = <j)gy/(x) defines an approximate action of
6 on H3. The proof then proceeds in five steps. The hard work, outlined in
Steps (1), (2), and (3), appears in other papers and is used to show that we may
assume that this approximate action extends naturally to the sphere at infinity
and that we may assume that the action on the sphere at infinity is conformai:
(1) The approximate action extends to an action of G on the 2-sphere at
infinity. An old argument of D. Mostow [M] shows that the extended action is
uniformly quasi-conformal on the 2-sphere.
(2) An argument of D. Sullivan [S] shows that after a quasi-conformal change
of coordinates at infinity, the action can be made conformai.
(3) A theorem of W. Thurston [T] or of Tukia and Väisälä [TV] shows that
the quasiconformal change of coordinates at infinity may be realized as the map
at infinity induced by a bi-Lipschitz homeomorphism of H3. This bi-Lipschitz
homeomorphism is of course a quasi-isometry of H3. That is, we may assume
that the original action at infinity was conformai.The conformai action at infinity induces an isometric action on hyperbolic
space.(4) The induced isometric action is properly discontinuous.
(5) The induced isometric action is cocompact.Steps (4) and (5) require some additional geometric argument which essentially
succeeds because the action of G on Y is both properly discontinuous and
cocompact, while, up to the factors that matter in the definitions of proper
discontinuity and cocompactness, the spaces Y and H3 are equivalent. Now
we give the details of the steps.
Step I. As noted in §1, the group G acts isometrically, properly discontinuously,
and cocompactly on Y by multiplication on the left. If tj> and \¡i are AT-quasi-
inverses as indicated above, then we may define an action of G on H3 by
the formula g(x) = <j>(g(i//(x))). Although this action is by relations (multi-
valued functions) rather than by continuous functions, nevertheless it extends
"continuously" in a unique way to the sphere S2^ at infinity for H3 and thisextension is quasiconformal on S2^ . Perhaps we should explain the meaning of
"continuous" extension in this setting. Let Xi, x2, ... denote a sequence from
License or copyright restrictions may apply to redistribution; see http://www.ams.org/journal-terms-of-use
424 J. W. CANNON AND DARYL COOPER
H3 converging to a point x of S__,. Then the sequence of sets <j)(g(y/(xi)))
converges to a unique point g(x) of S2^ . This induced action at infinity is
called the boundary action. The proof is essentially due to Mostow [M], where he
proved that a AT-quasi-isometry of H" to itself extends uniquely in the fashion
just mentioned to a A"2-quasiconformal map of S"_1. Though our definitions
are slightly different from Mostow's, Mostow's proof is easily adapted to our
situation.
Step 2. Now that G acts quasiconformally on S2^ with uniform constant of
quasi-conformality, a theorem of Sullivan [S] implies that this action is conju-
gate, by a quasiconformal map h : S2^ —► S___. , to a conformai group action ons2
Step 3. By a theorem of Thurston [T, Chapter 11] or of Tukia and Väisälä [TV,Theorem 3.2], the map h is the boundary map of a bi-Lipschitz, hence quasi-
isometric, homeomorphism of H3. Thus, replacing <j> if necessary by h o 0,
we may assume that our original quasi-isometry </> induced a conformai action
of G on S2^. Any conformai action of G on S2^ is the boundary action of
an isometric action of G on H3. Our original quasi-isometric action of G onH3 was induced by the quasi-isometry <f>. We shall denote the new isometricaction by O : G x H3 —► H3. It remains only to show that this isometric action
O of G on H3 is in fact both properly discontinuous and cocompact.
Step 4. We show that the action O is properly discontinuous.
Fix O e H3. Let a and ß denote perpendicular geodesic lines in H3 with
intersection a(~)ß = {0} . Let A and B denote graph geodesic lines in Y such
that 4>(A) shares infinite endpoints with A and 4>(B) shares infinite endpoints
with B . (See the Corollary from §2.) Let C denote a graph geodesic segmentjoining A to B, C of minimal possible length.
There is a positive number V(K), depending only on A", such that (p(C)
lies within V(K) of {0}. Indeed, there exist points a e A and b e B such
that <p(a) and <f>(b) lie within K + W(K) of O. Hence
dr<K(K+W(K)+l).
Hence C has length < A"(A- + W(K) + 1), and 4>(C) has
diameter < A^(diamC + 1).
But 4>(C) intersects both the W(K) neighborhood of a and the W(K) neigh-borhood of ß , and those neighborhoods diverge from one another as one moves
away from O. The existence of V(K) follows.
Suppose next that g e G. Fix P > 0 and suppose g e G satisfies
dH(0,g(0)) < P. We wish to estimate dT(C, gC). The sets g(a) andg(ß) are geodesies meeting perpendicularly at g(af]ß) = g(O), g(A) has im-
age (¡>g(A) which shares infinite endpoints with g(a), g(B) has image 4>g(G)
which shares infinite endpoints with g(ß), and g(C) is of minimal length
joining g(A) and g(B). Hence 4>g(C) lies within V(K) of g(0). Pickae C, be g(C). Let a' e <p(a), b' e <j)(b). Let a" e y/(a') Cf» (f>(a) and
License or copyright restrictions may apply to redistribution; see http://www.ams.org/journal-terms-of-use
HYPERBOLIC GROUPS 425
b" e p(b') cif/ o (¡)(b), where y/ is a A"-quasi-inverse to <j>. Then
dr(C,gC)<dr(a,b)
<[dr(a,a") + dr(b,b")] + dr(a",b")
<[2K] + K(dH(a',b') + l)
< 3A- + K[dH(a', O) + d„(b', g(O)) + d„(0, g(0))]
<3K + K[2V(K) + P].
The action of G on Y is properly discontinuous, so that only finitely many
elements of G can satisfy dT(C, gC) < 3A" + A"[2F(A") + P]. Hence all otherelements of G move O a distance > P. We conclude that G acts properly
discontinuously on H3.
Step 5. Finally we show that the action O of G on H3 is cocompact.
Let x G H3. Let x G y/(x). Let g e G be a vertex of Y within a distance1 of x'. We estimate the distance dn(0, g~xx) as follows.
dH(0,g-lx) = dH(gO,x)
< dH(gO, <f>g(Q) + dH(4>g(C), <ß(x')) + dH(<p(x'), x).
We have
dH(gO,<pg(C))<V(K),
dH(<f>gC, 0x') < KdT(gC, x') + K = K[dT(C, g-xx')] + K
< K(dr(C, id) + rfr(id, ,r'x')) + A"
< K(dr(C, id) + K) + K,
d(<f)x',x) <K,
the last since <j>x' e 4>v(x).We conclude that the closed ball or disk centered at O of radius
V(K) + [K(dr(C, id)+ K) + K] + K
contains a representative of each orbit of the action. Hence the action is co-compact.
4. The finite volume case
The difficulty in the finite volume case is illustrated by the following 2-
dimensional example. Consider an ideal triangle in hyperbolic 2-space H2.The group G generated by the three reflections in the sides of this triangle actsisometrically and properly discontinuously on H2 with finite 2-dimensional vol-
ume (= area). The images of the original triangle tile the hyperbolic plane. But
the action is not cocompact. The group G is isomorphic with the free product
of three copies of Z/2Z. The graph Y of G can be embedded in H2 : ver-tices of T go to an orbit of the action, namely to barycenters of the vertices of
the triangles, edges become hyperbolic geodesies joining barycenters of adjacenttriangles. If the action were cocompact, then this embedding would be a quasi-
isometry. However, if one draws the diagram, it is apparent that the embedded
graph completely misses infinitely many disjoint circular disks (horoballs) in H2
tangent to the circle at infinity. Each of these disks contains an infinite amountof area of H2 . That is, the graph is not really a good approximation to H2.
License or copyright restrictions may apply to redistribution; see http://www.ams.org/journal-terms-of-use
426 J. W. CANNON AND DARYL COOPER
In order to obtain a graph that is quasi-isometric with H2 , the graph must be
augmented by adding certain infinite portions that fill up the omitted horoballs.
In summary, in the finite volume, noncocompact case, we must augment the
Cayley graph if we are to obtain a graph quasi-isometric with hyperbolic space.The potential holes in the graph corresponding to omitted horoballs in hyper-
bolic space are represented by parabolic subgroups of the group. In dimension 3
parabolic subgroups can be easily recognized as groups having an Abelian group
of finite index, the Abelian group having rank 2. After one has identified the
potential holes and appropriately filled them, then the characterization merges
with that of the cocompact case. That is, in dimension 3, a group is finite-volume hyperbolic if and only if its augmented Cayley graph is quasi-isometric
with hyperbolic 3-space H3.Here is a brief outline of this section. The three steps will be carried out in
subsections 4.1, 4.2, and 4.3.(1) We define the augmented Cayley graph.
(2) We show that if a group is finite-volume hyperbolic in dimension 3, then
the augmented Cayley graph is quasi-isometric with hyperbolic 3-space H3.
(3) We show that if the augmented Cayley graph of a group is quasi-isometricwith H3, then the group is finite-volume hyperbolic in dimension 3.
4.1. The augmented Cayley graph. We wish to fill the parabolic holes in a
group. We need to be able to recognize those holes. We recall the structure
of a finite-volume hyperbolic group G in dimension 3 in order to see how
those parabolic holes arise. The structure of G is discussed, for example, in[Ma, Chapter VI]. Here are some standard properties of G and its parabolic
subgroups. The group G is finitely generated. Let x e S2^ be any point that is
fixed by some parabolic element of G. Let Gx denote the group of all elements
of G that fix x . Let Px denote the set of all parabolic elements of Gx togetherwith the identity element of G. Then Px is a free Abelian group of rank 2.
Every Abelian subgroup of G of rank > 1 lies in some such subgroup Px . The
group Gx is precisely the normalizer of Px in G, and the group Gx is also its
own normalizer in G. The group Px is of finite index in Gx ; consequently,
the group Gx is finitely generated. There are only finitely many subgroups Px
or Gx of G up to conjugacy. The group Px is called a parabolic subgroup of
G. Hence we have the following.
Necessary conditions for a group to be finite-volume hyperbolic in dimension 3. A
group G can be finite-volume hyperbolic in dimension 3 only if it satisfies the
following conditions. The group G is finitely generated. Any Abelian subgroup
of G of rank > 1 is a subgroup of a maximal Abelian subgroup which has rank
2. We call the maximal Abelian subgroups of G having rank 2 the parabolic
subgroups of G. Any parabolic subgroup of G has finite index in its normalizer,
and the normalizer is its own normalizer. There are only finitely many conjugacy
classes of parabolic subgroups of G.Given such a group, we construct the augmented Cayley graph Y+ as follows.
Let Px, ... , Pn denote a set of parabolic subgroups of G, one from each
of the finitely many conjugacy classes. Let G\,... ,Gn denote the normalizers
of the groups P. ,...,/*„. Let C = C~x denote a finite generating set for G
which includes, for each i, a finite generating set C, = C~ ' for each of the
License or copyright restrictions may apply to redistribution; see http://www.ams.org/journal-terms-of-use
HYPERBOLIC GROUPS 427
groups G i. We begin with the Cayley graph Y = Y(G, C). We metrize Y withthe word metric. This is the graph which we shall augment.
Let Gi denote one of the chosen normalizers. For each left coset gG¡ of G,
in G, let Y(gGj) denote the subgraph of Y whose vertices are the vertices of
gG¡ and whose edges are those labelled by the elements of C, which begin and
end at points of gG¡. Consider the graph of the semigroup gG, x {0, 1, 2, ...}
with generating set (C,, t), t generating Z+ = {0, 1,2,...} additively. Iden-
tify Y(gGi) x {0} with Y(gGi). The graph Y augmented by the collection ofsemigroup graphs just described, one for each left coset of each G,, is the
augmented group graph Y+ .
One does not assign length 1 to every edge of the augmented group graph. To
get an appropriate metric on Y+ one proceeds as follows. One assigns length 1to each edge of gGt x {0, 1, 2, ... } labelled by the generator t of Z+ . Butto any edge in a level gG¡ x {j} one assigns the length exp{-_/'} . This makes
r+ a metric space.Note that the group G acts on the augmented Cayley graph Y+ by multipli-
cation on the left. This fact has been noted before for the Cayley graph, andthe extension to the remainder of the graph is trivial.
4.2. Quasi-isometry.
Theorem. Let G denote a finite-volume hyperbolic group, and let Y+ denote
the augmented Cayley graph associated with G. Then the graph Y+ is quasi-
isometric with hyperbolic space H3.
Proof. We are given that G acts isometrically, properly discontinuously, and
with finite-volume on hyperbolic space H3. We describe a quasi-isometry
<b:Y+^U\
We first define </>|T. If v is a vertex of Y, then v e G and we define
(¡)(v) = v(0) G H3. If e = (v, c, v • c) is an edge, with v e G and c e C, then
(j) is to take e linearly onto the hyperbolic geodesic segment [4>(v), <f>(v • c)] =
[v(0),v(c(0))] joining v(0) and v(c(0)) in H3.
We next define, for each coset gG¡ of each G, G G, (j)\Y(gGi x Z+). Itsuffices to define the image of each vertex, for the edges are then to be taken
linearly onto the hyperbolic geodesic segments joining the images of the end-
points. Let Pi be the point at infinity fixed by the group G,. Then the point
gp¡ is the point at infinity fixed by the conjugate group gY¡g~x. Fix v e G¡.
Let [gv(0), gpi) denote the geodesic ray joining the point g(v(0)) = 4>(gv)
with the point g(p¡) G S2^ at infinity. Let Xo = g(v (0)), x\, x2,... denotepoints equally spaced on this ray with du(xo, x¡) = j. Define <f>(g'V, j) = Xj.
This completes the definition of <f>. It is clear that <j> is G-equivariant.
It remains to see that 0 is a quasi-isometry. The conditions that we shall
check are those of the following technical lemma whose proof we shall omit. We
remark only that the proof is essentially the same as the proof of [C, Theorem
1] or of the Equivalence Theorem of §1. (Recall that N(A, e) denotes the e
neighborhood of the set A.)
Lemma. Let X and Y be spaces with path metrics dx and dy ■ In order that
a function </>: X -> Y be a quasi-isometry it is sufficient that
(i) for some L > 0, re N(<f>(X), L) ;(ii) for some A" > 0 and for all xx,x2 G X, dY(4>(xi), <¡>(x2)) < K ■
dx(xi,x2); and
License or copyright restrictions may apply to redistribution; see http://www.ams.org/journal-terms-of-use
428 J. W. CANNON AND DARYL COOPER
(Hi) for each M > 0 there exists an N > 0 such that if dx(xi, x2) > N then
dY(<t>(xi), (f)(x2)) > M.
We now verify the three conditions of the lemma. It is helpful first to study
the structure of H3 relative to G.
The group G has a fundamental domain D with finitely many sides. Let
Pi, ... , pn be the points fixed by the groups Gi,..., G„ as above. Each of
these points is a translate of one of the cusps on the domain D. Let TV., ... , N„
denote disjoint open horoball neighborhoods of px, ... , p„ in H3. Require
that the G translates of the N¡ fill up precisely a family of disjoint open
horoballs in H3. The neighborhoods N¡ may be chosen so small that
^(r)cH3-UW:^G;i = l,...,fl}.
We call Q = H3 - \J{gNi : g G G ; /= 1,...,«} the cocompact part ofH3(rel G). This set Q has a compact intersection with the fundamental domain
D. This decomposition of H3(relG) into a compact and a horoball part can
be altered in an inessential way by adjusting the sizes of the TV,.
We now check condition (i) of the lemma. Let x G H3. Then either there
exists an i, I < i < n, and g e G such that x G gN¡ or there exists g e Gsuch that x G _?(D - (J TV,-). In the latter case,
dH(x, 4>Y+) < Diam g (ü - |J Ntj < oo.
In the former case, let B denote the horoball gN¡. A geodesic arc of length < 1takes x toward the center p = gp¡ of the horoball B and onto a horosphere
S centered at p which contains the 0-image of the vertices of some levelGp x {j} of the subgroup Gp of G fixing p . Every point of S is a uniformly
small distance 5 from a vertex of Gp x {j} , the size ô being independent of iand S. This observation completes the proof that condition (i) of the lemma
is satisfied with L > max{Diam g(D - (J N¡), ô + 1} .We now check condition (ii). Every edge of Y+ has either a label c e C or
a label í ora label t~x. Each edge with label t or t~l is mapped isometri-
cally by </>. An edge of Y+ with label c lies in a level Y(GP x {j}) and has
length l/eJ . The horospherical distance between the image endpoints is (l/eJ)times the horospherical distance between the image endpoints of an edge in Ywith label c. The ratio between hyperbolic distance and horospherical distance
is uniformly bounded since the hyperbolic distance is uniformly bounded. It
follows that the image edges have lengths whose ratios to preimage lengths areuniformly bounded. Condition (ii) follows immediately.
To check condition (iii) we assume M > 0 given. The neighborhoods N¿
of the cusps may be chosen so small that the only portion of Y+ mapped
into a horoball Bp centered at p and formed from translates of N¡ lies in
the associated augmentation Y(GP x Z+). We then consider the expansion
of the cocompact part Q to the closed hyperbolic (M + 1 )-neighborhoodcl(yV(ß, M + 1)) of Q in H3. Now let x., x2, G Y+ be given such that
dn(<j>(xx), <t>(x2)) < M. There are two cases to consider.
Case 1. Suppose (¡>(xx) and </>(x2) lie in cl(N(Q, M + 1)). If Y0 denotesthe portion of Y+ mapped into cl(N(Q, M + 1)) by <t>, then r/>|r0: r0 —►c\(N(Q, M + 1)) is a G-equivariant map between two spaces on which G acts
License or copyright restrictions may apply to redistribution; see http://www.ams.org/journal-terms-of-use
HYPERBOLIC GROUPS 429
cocompactly. Hence this map is a quasi-isometry. It follows that there is a
constant M' depending only on M and not on x. and x2 such that the
distance between xi and x2 in cl(N(Q, M + 1)) is less than or equal to M'.The M' neighborhood of a compact fundamental domain for
G\cl(N(Q,M+l)), cl(N(Q,M+l))
having its path metric, has compact closure. Hence the preimage of such aneighborhood has finite diameter D in Y+ , this diameter depending only on
the fundamental domain and M' and not on x. and x2 . Hence d(xx, x2) isuniformly bounded by D, independent of x_ and x2.
Case 2. Suppose 0(x.) and (¡>(x2) do not lie in cl(iV(ß, yV+1)). Then <j>(x\)
and (j>(x2) lie in a single horoball Bp as described above. The radial distance
(toward p) between levels of 4>(xx) and 4>(x2) is bounded by M. The horo-
spherical distance between projections into a single level near (¡>(xx) and 4>(x2)
is a bounded multiple of M. The horospherical graph distance is a bounded
multiple of horospherical distance. Hence graph distance is a bounded multipleof M.
Cases 1 and 2 complete the proof of (iii). Hence 0 is a quasi-isometry.
4.3. Recognizing a finite-volume group.
Theorem. Let G denote a group satisfying the necessary conditions of subsection
4.1. Let Y+ denote the augmented Cayley graph associated with G. If the
graph Y+ is quasi-isometric with hyperbolic space H3, then the group G is
finite-volume hyperbolic in dimension 3.
Proof. The proof proceeds in five steps, exactly as in the cocompact case.
Step 1. There are no changes from the cocompact case. Again, Mostow's argu-
ment applies. Hence there is an induced quasi-conformal action at infinity.
Step 2. There are no changes from the cocompact case. Again, Sullivan's ar-
gument applies. Hence after a quasi-conformal change of coordinates we mayassume that the action at infinity is conformai.
Step 3. There are no changes from the cocompact case. Again, Thurston's or
Tukia-Väisälä's argument applies. The quasi-conformal change of coordinates
at infinity may be realized by a bi-Lipschitz homeomorphism in H3. Hence we
lose no generality in assuming that the original action at infinity was conformai.
Step 4. There are no changes from the cocompact case. Again we conclude that
the induced isometric action on H3 is properly discontinuous. In particular,the kernel of the action is finite. It follows that the parabolic groups of G have
images under the action that are truly parabolic in H3.
Step 5. It remains only to show that the action is of finite-volume. To that end
we want to alter the quasi-isometry <f> so that it more nearly approximates the
function defined in the proof of the theorem in subsection 4.2. We need the
following lemma.
Lemma (the action of a parabolic quasi-isometry on horoballs). Suppose that
h : H3 —> H3 is a K-quasi-isometry, that h^ is the quasiconformal extension
of h to infinity, and that hoo is conformai and parabolic with fixed point x at
License or copyright restrictions may apply to redistribution; see http://www.ams.org/journal-terms-of-use
430 J. W. CANNON AND DARYL COOPER
infinity. Then h preserves horoballs at x within a fixed bound depending only
on K.
Proof of the horoball lemma. We may take as model for H3 the Poincaré up-
per half space with x at infinity. A point p on a horoball centered at x is
determined by the vertical ray R through p and by any geodesic S through
p orthogonal to R. The transformation h^ is then a Euclidean isometry of
the Euclidean plane S2X3\{x} which extends to a natural Euclidean isometry
of the upper half space model of H3 which we also denote by h^ , preserves
horoballs at x , takes vertical rays to vertical rays, and preserves the Euclidean
shape and size of all other geodesies. In particular the images of R and S un-
der hoo intersect at the h^ image of p . As noted earlier, the transformation
h preserves geodesies within a uniform bound depending only on K . It follows
that the image h(p) of p lies within a uniform bound of hoo(p), the bound
depending only on A".
We are now ready to complete the proof of the theorem. Fix one of the
parabolic subgroups G, of G. When we augmented the group graph along G,,
we added a family of parallel rays with edges labelled by the letter t. All ofthese rays are graph geodesies, hence their images in H3 converge to points at
infinity. Since the rays get closer and closer to each other toward infinity, they
converge to the same point x at infinity. Each of the transformations of G,
fixes x . Since G, has an Abelian subgroup of rank 2, and since the kernel of
the action is finite, the image of G, is a group of conformai transformations
fixing x and having an Abelian subgroup of rank 2 of finite index. Each of
the elements of the image must either be parabolic or elliptic, the parabolics
forming a subgroup of finite index. Of course, the elements of G do not act
directly on H3, but we lose no generality in assuming that <j> embeds Y+ on
which G does act faithfully. It is easy to extend the action discontinuously butquasi-isometrically to all of H3 by mapping each point of H3 to a nearby point
of the image of (¡>. The lemma then applies and shows that the augmentation
part of the graph is A"-dense in some horoball at x . In particular the remainder
of the graph cannot really intrude on that horoball after a certain point. What
that allows one to do is to make the mapping <f> truly G equivariant on thehoroball near x without losing the property of quasi-isometry. One can change
the map in exactly the same way in each G translate of that horoball, and one
can deal similarly with the other groups G¡.
With (p altered as in the preceding paragraph, one can return to the proof of
§3 which showed the cocompactness of the group G of §3. The same argument
shows that the G of §4 acts cocompactly off of the set of horoballs just dealt
with. But there are only finitely many G orbits of such horoballs, and the
quotient of each has finite volume. We conclude that G is of finite volume.
References
[BK] P. Buser and H. Karcher, Gromov's almost flat manifolds, Astérisque 81 (1981).
[Cl] J. W. Cannon, The combinatorial structure of cocompact discrete hyperbolic groups, Geom.
Dedicata 16 (1984), 123-148.
[C2] _, The theory of negatively curved spaces and groups, Ergodic Theory, Symbolic Dynam-
ics and Hyperbolic Spaces (Tim Bedford, Michael Keane and Caroline Series, eds.), Oxford
Univ. Press, Oxford, New York and Tokyo, 1991, pp. 315-369.
License or copyright restrictions may apply to redistribution; see http://www.ams.org/journal-terms-of-use
HYPERBOLIC GROUPS 431
[F] W. J. Floyd, Group completions and limit sets ofKleinian groups, Invent. Math. 57 (1980),
205-218.
[Gl] M. Gromov, Hyperbolic manifolds, Groups and Actions, Proc. Stony Brook Conf., Ann. of
Math. Studies No. 97, Princeton Univ. Press, Princeton, N.J., 1981, pp. 183-215.
[G2] _, Infinite groups as geometric objects, Proc. Internat. Congr. Math., Elsevier, New York,
1983, pp. 385-392.
[G3] _, Groups of polynomial growth and expanding maps, Inst. Hautes Études Sei. Publ.
Math. 53(1981), 53-73.
[Ma] Bernard Maskit, Kleinian groups, Springer-Verlag, Berlin, Heidelberg and New York, 1988.
[M] G. D. Mostow, Strong rigidity of locally symmetric spaces, Ann. of Math. Studies No. 78,
Princeton Univ. Press, N.J., 1972.
[S] D. Sullivan, On the ergodic theory at infinity of an arbitrary discrete group of hyperbolic
motions, Proc. Stony Brook Conf., Ann. of Math. Studies No. 97, Princeton Univ. Press,
Princeton, N.J., 1981, pp. 465-496.
[T] W. P. Thurston, The geometry and topology of 3-manifolds, Lecture Notes, Princeton Univ.,
1978.
[TV] P. Tukia and J. Väisälä, Quasiconformal extension from dimension n to n + 1 , Ann. of
Math. (2) 115(1982), 331-348.
Department of Mathematics, Brigham Young University, Provo, Utah 84602
Department of Mathematics, Warwick University, Coventry CV4 7AL, England
Current address: Department of Mathematics, University of California at Santa Barbara, Santa
Barbara, California 93107
License or copyright restrictions may apply to redistribution; see http://www.ams.org/journal-terms-of-use